基于神经网络GIS局部放电模式的识别  

Neural Network Based Partial Discharge Pattern Identification of GIS

作　　者：李培江[1,2] 朱晓锦[1] 尤婷[3] 

机构地区：[1]上海大学机电工程与自动化学院,上海200072 [2]衢州职业技术学院,浙江衢州324000 [3]衢州学院,浙江衢州324000

出　　处：《中国电力》2013年第10期60-66,83,共8页Electric Power

基　　金：浙江省自然科学基金资助项目(Y1110557)

摘　　要：全封闭气体绝缘开关设备(GIS)广泛应用于电网中,其内部缺陷导致的设备故障可能会引起大面积停电事故。针对GIS缺陷放电模式识别问题,设计了3种GIS典型放电模式,通过实验平台获取放电指纹数据,并从中提取出12种特征。对基于单一网络方式的概率神经网络、自适应神经网络以及基于复合神经网络方式下的GIS局部放电识别问题进行对比研究,考察3种网络方式在输入验证、部分训练集等不同条件下的放电模式识别率与一致性问题。实验结果表明,采用上述单一方式神经网络可以作为一种局部放电识别手段,但识别结果的一致性较差,而复合神经网络不仅具有高识别率,而且一致性也较好,可以较好地满足GIS局部放电识别。Gas insulated switchgears （GIS） have been widely used in power grids.Equipment faults cased by internal defects of GIS,however,may cause large-area outage.For identification of discharge patterns of GIS defects,three typical GIS discharge patterns are designed in this paper.The fingerprint data are obtained by testing platform and 12 kinds of features are extracted from these data.A comparative study is conducted on GIS partial discharge pattern identification based on probabilistic neural network and adaptive neural network.The issue about input validation,partial training sets and composite neural network are tested in sequence.The test result shows that the probabilistic neural network and adaptive neural network can be used as an effective means of discharge pattern identification.Moreover,the composite neural network has a good identification consistency and high identification efficiency.

关 键 词：GIS 局部放电 神经网络 模式识别 

分 类 号：TM835[电气工程—高电压与绝缘技术]